www.jpct.net

General => News => Topic started by: EgonOlsen on October 01, 2009, 08:13:00 pm

Title: Software renderer performance
Post by: EgonOlsen on October 01, 2009, 08:13:00 pm

I did some simple benchmarks with the multithreaded software renderer in 1.19.

All benchmarks were done on a Core2 Quad @ 3.2Ghz.

Raw fillrate:

1 Core:  40 Megapixel/sec.
2 Cores: 77 Megapixel/sec.
3 Cores: 94 Megapixel/sec.
4 Cores: 110 Megapixel/sec.

Real world performance, Quake 3 level:

1 Core:  35 fps
2 Cores: 60 fps
3 Cores: 80 fps
4 Cores: 90 fps

If all goes well, i'll release 1.19 within the next week.

Title: Re: Software renderer performance
Post by: JavaMan on October 01, 2009, 09:07:04 pm

Looking awesome! Dang if I didn't have those blasted midterms.  >:( >:( >:( >:( >:( >:( >:( >:( >:( >:( >:( >:(

Title: Re: Software renderer performance
Post by: raft on October 02, 2009, 01:47:11 pm

sounds great :) so how do you achieve this ? are you making an architectural change like going multi-threaded ?

Title: Re: Software renderer performance
Post by: raft on October 02, 2009, 01:58:15 pm

i've got the answer in: http://www.jpct.net/forum2/index.php/topic,1435.0.html
jPCT goes multi-threaded :)

Title: Re: Software renderer performance
Post by: EgonOlsen on October 02, 2009, 03:21:11 pm

The hardware renderer could make use of mt anyway, but it's limited to two threads. I tried to make the software renderer use mt multiple times, but never got the desired results. This time, i got it right...at least i think i did. It will scale up to an unlimited number of cores, but i don't think that anything higher than 8 cores will improve performance much.